The gfp-ohyA gene was constructed to contain an amino terminal His₆ tag followed by the gfp gene that was linked by a flexible linker (amino acid sequence GGGGSGGSS) to ohyA as previously described (Radka et al., 2024). The initiating methionine residues of GFP and OhyA were mutated to alanine residues to eliminate alternative transcription start sites during overexpression. Thus, the open reading frame encoded fMet-His₆ tag-GFP-flexible linker-OhyA. The gfp-ohyA gene was cloned into the pET28a 5′-NcoI restriction site using the Gibson Assembly method to make pGFP-OhyA. Overexpression of gfp-ohyA is driven by the strong bacteriophage T7 promoter in this plasmid.

The pGFP-OhyA plasmid was transformed into Escherichia coli BL21 (DE3) cells, and isolates were selected on Luria broth agar plates containing 50 μg/μL kanamycin (Gold Biotechnology). The transformants were cultured in Luria broth with 50 μg/μL kanamycin at 37°C with shaking at 200 rpm. Once the cells reached an OD₆₀₀ of 0.6, the temperature was lowered to 16°C, and protein expression was induced overnight with 1 mM isopropyl-β-D-thiogalactoside (Gold Biotechnology). After induction, the cells were harvested and lysed in a buffer containing 20 mM Tris, pH 8.0, 10 mM imidazole, 200 mM NaCl. The GFP-OhyA protein was purified from the lysate using nickel agarose beads (Gold Biotechnology) and eluted with a buffer containing 20 mM Tris, pH 8.0, 250 mM imidazole, and 200 mM NaCl. The eluted protein was further purified by gel filtration using a HiLoad Superdex 200 16/60 column (Cytiva Life Sciences) into a buffer with 20 mM Tris, pH 8.0, and 200 mM NaCl. The molecular weight of GFP-OhyA was estimated by SEC-MALS using a Superose 6 Increase 10/300 GL column (Cytiva Life Sciences) with three detectors connected in series: an Agilent 1200 UV detector (Agilent Technologies), a Wyatt DAN HELEOS II multi-angle light-scattering and a Wyatt Optilab T-rEX differential refractive index detector (Wyatt Technologies). The column was equilibrated in 20 mM Tris, 200 mM NaCl, pH 7.6, and 100 μL of GFP-OhyA (2 mg/mL) was injected at a flow rate of 0.5 mL/min at 25°C. Data were recorded and analyzed with the Wyatt Astra software (version 8), and plotted as a molar mass distribution superimposed on a chromatogram of A₂₈₀ versus elution volume.

The thermal stability of GFP-OhyA was determined by a Sypro Orange-based fluorescence assay (Huynh and Partch, 2015). Solutions (30 μL) of GFP-OhyA (1 mg/mL) in 50 mM K₂HPO₄, 150 mM NaCl, pH 6, and 2.5 × Sypro Orange dye were added to the wells of a ThermoGrid optically clear PCR plate (Denville Scientific). The plates were centrifuged at 1,000 × g for 5 min and then analyzed by the ABI 7300 real-time PCR system as described previously (Radka et al., 2020). The temperature was ramped from 25°C to 95°C at 1°C/min with the fluorescence read six times at each temperature ramp. The resulting data were fitted to a Boltzmann sigmoidal equation to determine the melting point. The experiment was repeated five times, the thermal melting temperature of each replicate was determined independently, and the melting points from each replicate were averaged to determine the reported thermal melting point.

Charged unilamellar vesicles were prepared from three different lipid compositions: 100% anionic dioleoylphosphatidylglycerol (DOPG) (Avanti Polar Lipids), cationic dioleoylphosphatidylcholine (DOPC) (Avanti Polar Lipids), and a 99:1 mixture of DOPG and 18:1Δ9 (Ambeed) (DOPG/OA). For 100% DOPG or 100% DOPC vesicles, 60 μg of lipid was dissolved in chloroform and added to a 16 mm glass culture tube. For the DOPG/OA vesicles, 60 μg of DOPG and 2.12 μg of 18:1Δ9 were added to the glass culture tube. Micelles were prepared from 60 μg neutral triolein (Avanti Polar Lipids). The chloroform was evaporated under nitrogen gas using the Reacti-VapIII (Thermo Scientific, #TS-18826). The dried lipids were resuspended to a final concentration of 3 mM lipids in 50 mM KPO₄, 150 mM NaCl, pH 6.0 and sonicated at 37^∘C for at 5–10 min. Lipids were then manually extruded back and forth 75 times using a 10 mm filter support (Avanti Polar Lipids) fitted with either 0.03 μm or 0.1 μm polycarbonate membranes to produce SUVs or LUVs, respectively. The 0.1 μm polycarbonate membrane was used to prepare micelles. The hydrodynamic size and polydispersity of each lipid particle were measured using dynamic light scattering (DLS) with the RNA-LNP application of a Stunner instrument (Unchained Labs). Measurements were conducted under default settings: a 142° scattering angle, a 660 nm laser, and four 1 s acquisitions with automatic angle selection and outlier exclusion. At a lipid concentration of 3 μM, DLS confirmed that lipid particles remained monodisperse and did not aggregate, even at the highest lipid concentration used in the binding titration.

To fluorescently label the vesicles or micelles for FCS standardization, 9.76 μL of a 1:100 dilution from a 1 mg/mL 18:1 Liss Rhod PE (Avanti Polar Lipids) stock solution was added to each lipid mixture in a 16 mm glass culture tube. The chloroform was evaporated under nitrogen gas, and the vesicles or micelles were then assembled as described above. This process produced Rhod PE-labeled SUVs, LUVs, and micelles for assessing free diffusion.

To determine the impact of protein concentration on binding, 0.014, 0.028, 0.056, 0.113, 0.22, 0.45, 0.9, 1.8 μM GFP-OhyA was mixed with 0.188 μM accessible lipid. To determine the impact of accessible lipid concentration on binding, 0.0115, 0.023, 0.047, 0.094, 0.188, 0.375, 0.75, 1.5, 3 μM lipid was mixed with 0.113 μM GFP-OhyA. The samples were 50 μL in volume and incubated in 50 mM KPO₄, 150 mM NaCl, pH 6.0 for 5 m at 20°C before FCS data acquisition. The samples used for FCS standardization were 0.113 μM GFP-OhyA and 0.188 μM Rhod PE-labeled DOPC LUVs, DOPG SUVs and LUVs, DOPG/OA SUVs and LUVs, and triolein micelles.

FCS measurements were conducted using a custom-built microscope within the University of Kentucky Microscopy Core. Approximately 50 μL of freshly prepared, fluorescently labeled sample was placed onto an 18 × 18 mm Zeiss no. 1.5 coverslip, positioned inside a 35 mm Petri dish with an 18 mm hole (Cell E&G, Cat #: PDH00002-200). The Petri dish was mounted on a Nikon Eclipse Ti2 microscope equipped with a PicoQuant PicoHarp 300 Time-Correlated Single Photon Counting (TCSPC) system. A 532 nm laser, set at 1% power (29 μW), was used to excite the fluorescent labels, and a ×60 water immersion objective focused the laser beam on the sample. Photon detection was handled by the PicoHarp 300 TCSPC module. Each FCS result represents the average of three 60 s measurements. To avoid background interference from immobilized molecules on the coverslip, all measurements were taken at least 30 μm above the glass surface. Raw data was processed using SymPhoTime 64 (PicoQuant) software.

Laser power and pinhole diameter were calibrated using highly diluted GFP-OhyA and Rhod PE solutions, excited at 488 nm and 563 nm, respectively. At least five autocorrelation functions were obtained at 20°C for each condition, adjusting the laser irradiance to achieve consistent molecular brightness at 1% power. The measured laser power was 29 μW, and pinhole calibration ensured that the number of fluorescent molecules within the confocal plane (N) equaled 1.

All raw data analysis was performed through SymPhoTime 64. After recording the fluctuations of the times versus fluorescence intensity trace, which indicated the diffusing fluorescent species in the detection volume, the autocorrelation function is applied and defined as: G τ = < I t ⋅ I t + τ > < I t > 2 (1)

The parameter I (t) represents the intensity time trace (Hz). The brackets indicate averaging over time. The autocorrelation time refers to the total duration during which the fluorescent species remains within the confocal plane.

For our standard measurements, the autocorrelation data was further fitted with a pure diffusion model: G t = ∑ i = 0 n D i f f − 1   ρ i 1 + t τ D i f f i 1 + t τ D i f f i k 2 (2)

The parameter τ D i f f is the diffusion of the ith diffusing species in ms, n D i f f represents the number of independently diffusing species, ρ represents the contribution of the ith diffusing species, and κ is the length of diameter ratio of the focal volume. This fitting model allows us to extract the diffusion time and number of molecules for each of our standard measurements, assuming only diffusion contributions are present in solution.

For our experimental data, GFP-OhyA plus lipid particles, the autocorrelation data was further fitted with a 3D-triplet kinetics model comprising 2 components: G t = 1 + ∑ j = 0 n T r i p − 1 T j exp − t τ T r i p j ∑ i = 0 n D i f f − 1 ρ i 1 + t τ D i f f i α i 1 + t τ D i f f i α i κ 2 0.5 + G Inf (3)

The parameter τ T r i p is the lifetime of the dark (triplet) state and η T r i p is the number of dark (triplet) states. A triplet state refers to the non-fluorescent, long-lived excitation that causes a transition to the normal excited state. This event leads to a temporary “flickering” of fluorescence due to its dark nature. The presence of a triplet state can impact the shape of the ACF curve by causing a dip at the start of the curve. The anomaly parameter (α) of the ith diffusing species describes data points that significantly deviate from the expected pattern, thus enabling exclusion of these measurements when generating the model fit. G Inf corresponds to the correlation offset. This fitting model describes diffusion of our species with triplet state blinking. This allows us to extract the diffusion time and number of molecules for each of our diffusing species by appropriately describing the triplet state and hindered diffusion.

We employed the single-molecule FCS technique to measure how vesicle addition affects protein diffusion, specifically by tracking the protein molecules rather than the vesicles themselves. The observed shift in protein diffusion time is caused by specific binding, as demonstrated by the side-by-side comparison of multiple lipid types (Section 3.2). The key readout from this experiment is the proportion of protein molecules that diffuse freely versus those with inhibited diffusion, which we interpret as the fraction of protein bound. It is important to note that we did not measure the number of protein molecules on each vesicle or the extent of the vesicle surface area covered.

DOPG (12 mM) was prepared in a buffer containing 150 mM NaCl, 1 mM bis-tris methane, pH 6.0, and 10% D₂O. GFP-OhyA was titrated into the DOPG solution at varying concentrations (3.9, 7.5, 16.7, 28.1, 42.9, 82.3, and 169.9 μM), using identical buffer conditions as the DOPG preparation.

³¹P NMR experiments were conducted at 298 K using a Bruker Avance 400 MHz spectrometer operating at 161.7 MHz. Proton decoupling was achieved with inverse-gated decoupling to minimize ¹H–³¹P nuclear Overhauser effect (NOE) interactions. Data were collected using an observation frequency range of 65,789.48 Hz and a total of 64,000 data points. Acquisition parameters included a 90° pulse length and a 2 s pulse cycle. Approximately 22,000 accumulations were recorded before performing Fourier transformation of the free induction decay signal.

Two-dimensional ¹H-¹³C spectra were acquired to analyze interactions between DOPG and GFP-OhyA. DOPG (12 mM) was prepared in a buffer containing 150 mM NaCl, 1 mM bis-tris methane, pH 6.0, and 10% D₂O, with and without 169.9 μM GFP-OhyA. Spectra were recorded at 298 K using a Bruker Avance 600 MHz NMR spectrometer equipped with four RF channels and a 5 mm z-gradient TXI CryoProbe (Bruker BioSpin), and operated with a ¹³C detection frequency of 150 MHz. This setup allowed for detailed analysis of ¹H-¹³C correlations to investigate molecular interactions between the lipid and protein under defined conditions.

The χ² values were used to assess the quality of fit for FCS data modeling in SymPhoTime 64. Statistical analyses and mathematical modeling of the processed data were conducted using GraphPad Prism software version 10.3.0.

The specific binding with Hill slope model is defined by the equation: Y = B max * X h K D h + X h (4)

The binding model describes the relationship between ligand concentration (X) and specific binding (Y), with key parameters providing insights into the binding process. X represents the ligand concentration, while Y denotes the specific binding. The maximum binding capacity, B _max , reflects the total number of available binding sites, and the dissociation constant, K _D , indicates the ligand concentration required to achieve half-maximum binding at equilibrium, representing the binding affinity. The Hill slope (h) provides information about cooperativity: when h = 1, there is no cooperativity, meaning ligand binding at one site does not influence others. A Hill slope h > 1 indicates positive cooperativity, where binding at one site enhances the likelihood of ligand binding at additional sites, while h < 1 suggests negative cooperativity, where binding at one site reduces the affinity of other sites or indicates the presence of multiple binding sites with differing affinities.

An amino-terminal His-tagged version of GFP-OhyA (Figure 1A) was expressed in Eschericia coli and purified by Ni⁺ affinity and gel filtration chromatography to obtain a homogenous 97 kDa protein (Figure 1B). GFP-OhyA eluted predominantly as a single species on a Superose 6 Increase 10/300 GL column (Figure 1C). Like OhyA, the apparent molecular weight of GFP-OhyA estimated by its anisotropic light scattering pattern indicated the protein is 80.6% dimer in solution (Figure 1C). The 46.99°C ± 0.29°C melting temperature of GFP-OhyA (Figure 1D) indicates its structural resistance to thermal denaturation is within 1.31°C of OhyA (Radka et al., 2024). These data indicate GFP-OhyA is a stable protein with the correct OhyA quaternary structure.

Fluorescence correlation spectroscopy (FCS) was used to monitor the binding of fluorescently labeled ligands to biomolecules (Figure 2A). Diffusion time (τ_D), which scales with the hydrodynamic radius of diffusing molecules, was calculated from the autocorrelation of fluorescence time traces. We first compared the diffusion times of GFP-OhyA and Rhodamine PE-labelled large unilamellar vesicles (LUVs) and micelles. Fluorescence bursts in the 60 s time trace correspond to molecules passing through the confocal plane (Figure 2B). The intensity and duration of these bursts reflect fluorophore concentration and residence time in the detection volume.

Autocorrelation functions (ACF) derived from fluorescence time traces were used to determine diffusion times and molecular concentrations (Equations 1, 2). The best-fit normalized ACFs, G(τ), for GFP-OhyA is shown in Figure 2C, with a diffusion time of 0.032 ± 0.0012 ms. To evaluate lipid-binding preferences, we constructed LUVs of dioleoylphosphatidylglycerol (DOPG, anionic), dioleoylphosphatidylcholine (DOPC, cationic), and micelles of triolein (neutral lipid). Best fit ACFs for LUVs or micelles (Figures 3A–C; Table 1) exhibited longer diffusion times compared to free GFP-OhyA: 2.91 ± 0.31 ms for DOPG LUVs, 2.6 ± 0.27 ms for DOPC LUVs, and 0.65 ± 0.09 ms for triolein micelles. Dynamic light scattering measurements confirmed the lipid particles were monodisperse and of the expected size (Figure 3D). These results confirm that FCS effectively distinguishes free GFP-OhyA from lipid particle-bound protein based on shifts in diffusion time.

When GFP-OhyA binds its 18:1Δ9 (OA) substrate embedded in lipid membranes, its diffusion time increases–a hallmark of ligand binding to larger biomolecular complexes. Given the larger diameter of lipid particles relative to GFP-OhyA, we assumed that protein-loaded lipid particles diffuse with a similar coefficient as free particles. Mixing GFP-OhyA with lipid particles revealed a notable increase in diffusion time with DOPG LUVs but not DOPC LUVs or triolein micelles (Figure 4). Upon binding DOPG LUVs, GFP-OhyA diffusion time increased to 1.2 ± 0.12 ms and the ACF curve shape changed significantly Equation 3. Specifically, the upper plateau shifted downward, the lower plateau shifted rightward, and the slope connecting the plateaus broadened. These shape changes, combined with the shift to longer diffusion times, indicate that GFP-OhyA binds and interacts preferentially with anionic phospholipids.

To investigate the effects of substrate availability and membrane curvature on GFP-OhyA binding, we prepared DOPG/OA LUVs and small unilamellar vesicles (SUVs) of DOPG and DOPG/OA (Figures 5A–D). Mixing GFP-OhyA with DOPG/OA LUVs increased its diffusion time to 2.6 ± 0.22 ms. With SUVs, diffusion times were 1.8 ± 0.42 ms for DOPG and 2.2 ± 0.14 ms for DOPG/OA Equation 3. While all DOPG-containing vesicles shifted the lower ACF plateau rightward and broadened the slope connecting the plateaus, only DOPG/OA LUVs caused a downward shift of the upper plateau (Figure 6). These findings indicate that GFP-OhyA interacts with all DOPG-containing vesicles, with larger LUVs exerting a more pronounced effect on the ACF curve than smaller SUVs. Although embedding OA substrates into vesicles had minimal impact on the substrate-free vesicle ACF curve’s shape, it consistently extended GFP-OhyA diffusion times.

To further investigate the concentration-dependent binding of GFP-OhyA to vesicles, we used the SymPhoTime-derived ρ1 and ρ2 parameters from the fitted ACFs, reflecting the contributions of free and bound GFP-OhyA, respectively. The linear relationship between the GFP-OhyA concentrations used in the titration and those detected provides evidence that our measurements accurately correlate with the concentrations applied in each instance (Supplementary Figure S1). The fraction of GFP-OhyA bound was calculated and fitted to a specific binding with Hill slope model (Figure 7) (Equation 4). This binding model enables the extraction of the ligand concentration required to achieve half-maximum binding at equilibrium and fits a Hill slope to measure cooperativity. Results showed that substrate presence and lower membrane curvature influenced binding. At the highest GFP-OhyA concentration, the fraction bound to DOPG vesicles was 0.335 for SUVs (Figure 7A) and 0.442 for LUVs (Figure 7B). When DOPG/OA vesicles were used, these fractions increased to 0.412 for SUVs (Figure 7C) and 0.718 for LUVs (Figure 7D). Positive cooperativity was observed only when GFP-OhyA bound to substrate-free SUVs; in all other cases, the cooperativity was negative. These findings indicate that in both SUVs and LUVs, the presence of embedded substrate enhanced binding compared to substrate-free vesicles, despite reducing the binding affinity by 3- to 4-fold (from 79.9 to 247.4 nM for LUVs, and 117.1–447.0 nM for SUVs). The incomplete binding at high concentrations (fraction bound not reaching 1 asymptotically) suggests a reversible process with potential conformational states preventing complete vesicle saturation.

The influence of accessible lipid concentration on GFP-OhyA binding was analyzed similarly to the concentration-dependent experiments. A specific binding with Hill slope model was used to fit the protein fraction bound (Figure 8). Interestingly, as lipid concentration increased, GFP-OhyA binding decreased. For DOPG vesicles, the SUVs initially increased in bound fraction before decreasing from 0.618 to 0.170 (Figure 8A), while the LUVs showed a gradual decline in bound fraction from 0.547 to 0.289 (Figure 8B). In DOPG/OA vesicles, the SUVs fraction bound decreased from 0.444 to 0.157 (Figure 8C), and the LUVs bound fraction decreased from 0.573 to 0.217 (Figure 8D). Negative cooperativity was observed across all vesicle types. These results suggest that higher lipid concentrations may impede GFP-OhyA binding.

We normalized our results by calculating the ratio of bound GFP-OhyA to accessible lipid concentration to understand the role of protein distribution in membrane binding. As protein concentration increases with a fixed accessible lipid concentration, the ratio of vesicle-bound protein to accessible lipid concentration rises hyperbolically, demonstrating positive cooperative binding across all DOPG vesicle types (Figure 9A). This curve shape, with a distinct maximum, indicates that the number of vesicles carrying large quantities of protein increases relative to the protein molecules in solution. This suggests a protein:lipid molar ratio-dependent binding mode where protein molecules assemble on individual vesicles.

Conversely, increasing the lipid concentration while maintaining a constant protein concentration results is an exponential decline in the ratio of vesicle-bound protein to accessible lipid concentration, indicating negative cooperative binding across all DOPG vesicle types (Figure 9B). This curve, which exhibits a noticeable minimum, reflects a shift in the equilibrium between bound and unbound protein. As vesicle concentration increases, the protein-to-lipid molar ratio decreases, and the equilibrium shifts to a state with minimal protein bound, as the binding remains highly reversible. The apparent reduction in bound protein is not due to vesicles rendering bound protein undetectable by FCS, but rather to the equilibrium favoring the unbound state. This binding mode involves protein adsorption to the bilayer through electrostatic interactions, potentially disrupting protein assemblies or inhibiting their formation due to limited cooperativity or steric hindrance. Some protein molecules may bind to lipid sites in a way that excludes additional GFP-OhyA subunits from assembling, leaving unincorporated subunits unbound at fixed accessible lipid concentrations. This behavior underscores the dynamic and non-uniform nature of OhyA’s interaction with lipid membranes.

Notably, a membrane bilayer is essential for transitioning OhyA from discrete dimers to oligomeric ring assemblies that encircle vesicles (Radka et al., 2024; Oldham et al., 2024). The near-identical convergence of curves from both titration experiments indicates that OhyA’s association with the membrane is not significantly influenced by membrane curvature or substrate availability. Instead, protein-membrane binding is predominantly driven by protein:lipid molar ratio, reflecting the interplay between protein-protein and protein-lipid interactions.

Phosphorus-31 (³¹P) NMR spectroscopy was used to investigate the molecular interactions between GFP-OhyA and DOPG. Stacked NMR spectra (Figure 10A) revealed uniform and significant chemical shift changes (0.24 ppm) as the protein concentration increased, transitioning from the unbound state (0.68 ppm) to the bound state (2.14 ppm). These observations indicate fast-exchange binding of GFP-OhyA to DOPG that forms a slightly different size vesicle, characterized by a single sharp peak progressively shifting downfield with increasing protein concentrations.

At a protein concentration of 7.5 μM, a second smaller, broader peak appeared at 0.24 ppm, intensifying and broadening with increasing protein concentrations. This second peak, which shifts upfield of the unbound DOPG signal, remains at a constant chemical shift position and is consistent with intermediate-exchange binding with another vesicle of the same order.

The presence of both fast- and intermediate-exchange binding suggests two distinct binding events occur between GFP-OhyA and DOPG. At a protein-to-lipid concentration ratio slightly less than 1:50, the fast-exchange binding equilibrates, with no further chemical shift changes observed, while the intermediate-exchange peak intensity and broadening reaches its maximum. Two-dimensional ¹H–¹³C HSQC NMR spectra were collected for DOPG LUVs in the presence and absence of GFP-OhyA. The hydrocarbon peaks of the lipids exhibited significant overlap in both conditions, indicating that the lipid structure and conformation remain unchanged upon protein binding (Supplementary Figure S2).

The dissociation constant (K_D) for the fast-exchange binding was determined to be 3.181 ± 1.272 μM (Figure 10B). These results suggest that at low concentrations, GFP-OhyA transiently binds and releases from the membrane bilayer (fast exchange). In contrast, at higher protein concentrations, GFP-OhyA forms longer-lasting interactions on the bilayer as the peak intensity decreased, likely as part of an oligomeric complex.

We propose that OhyA exhibits cooperative switching (Duke et al., 2001), where the molecules dynamically transition between individual molecules and oligomers in response to the concentration of nearby membrane-bound proteins. This transition is driven by conformational changes that promote oligomer formation through interactions with other OhyA molecules on the same membrane bilayer. Oligomer assembly is stabilized by membrane binding and destabilized by dissociation. OhyA stochastically switches between these states, with the likelihood of each state dependent on the local concentration of membrane-bound proteins. Membrane binding shifts the equilibrium toward the oligomeric state, as coupling energy from protein nucleation on the bilayer induces conformational changes propagated through allosteric interactions.

The occupancy of membrane-binding sites by individual molecules is a nonlinear function of protein concentration, as membrane affinity increases with the amount of bound protein (Stefan and Le Novere, 2013). Small changes in lipid concentration can disrupt oligomer assembly by increasing the fraction of unbound protein, and separating individual molecules. Under these conditions, transient oligomers form and rapidly dissociate. At higher concentrations of membrane-bound proteins, stable oligomers predominate. At intermediate concentrations, the system exhibits bistability, oscillating between oligomeric assemblies and dispersed molecules. This model underscores the critical roles of membrane binding and cooperative protein-protein interactions in regulating OhyA’s oligomeric switch, driven by conformation-dependent interactions between adjacent protomers.